UI Reference: Application runtime settings - Red Hat Ansible Automation Platform

AI / LLM tab

These settings control large language model behavior in agentic workflows.

Agentic max completion tokens: Optional cap on the number of tokens the LLM may generate in a single agentic response. Set to 0 to use the provider's model-specific default (recommended). Set a positive value to enforce a hard output-length limit. **Default:** 0 (no limit).

System tab

System log level

Determines how much detail the system records. Changes are applied dynamically without a restart. Logging levels follow a hierarchical threshold. When you set a level, the application records everything at that level and above. INFO or WARNING is standard for production. Use DEBUG only during troubleshooting, as it creates large volumes of data and can slow down the application. Values: DEBUG, INFO, WARNING, ERROR, CRITICAL.

Performance test mode

Activates the in-memory metrics store and exposes internal metrics endpoints for performance testing.

Enabled: Provides unparalleled visibility into tail latency (the slowest 1% of requests) and intermittent bottlenecks. Raw metric records are stored in memory and queryable without restarting the application. Running heavy loads can cause high memory usage. Use this mode only for short-term testing in non-production environments, not continuous operation.
Disabled: The application runs with minimal overhead, only emitting summarized heartbeats or logs. This is the safe state for production environments.

Context Manager tab

Compression

Compression loop

Number of iterative compression passes allowed when the context still exceeds the token budget after the first pass. Each additional pass further reduces the content. Set to 0 (minimum) to disable iterative compression.

Compression max tokens

Token limit for the LLM response during abstractive compression. Controls the maximum length of the compressed output. Use a value smaller than the original context to achieve meaningful reduction.

Compression mode

How the system shrinks the retrieved text to fit into the token limit.

Extractive: Pulls exact sentences. This is safer as there is no hallucination risk.
Abstractive: Uses an LLM to summarize the snippets. This method is more coherent, but uses more processing time and carries a slight hallucination risk.

Compression temperature

LLM sampling temperature used during abstractive compression. Lower values (0.0-0.3) produce more deterministic, faithful summaries, which is best for factual extraction. Higher values (0.7-1.0) increase variety but significantly raise the risk of hallucinations. Only applies when compression mode is set to abstractive.

Retrieval (RAG)

Context Manager Default K

The number of document chunks or snippets to retrieve from the database.

Tradeoffs:
- High K (for example, 20+): Better retrieval in that you are more likely to find the answer. However, it introduces noise, which can confuse the LLM and increase costs.
- Low K (for example, 3-5): Reduced noise and lower cost, but the system might miss the relevant information if it is not in the top few results.

Hybrid search

Combines semantic (embedding-based) and lexical (keyword-based) search for document retrieval. Semantic search finds meaning, while lexical search finds exact keyword matches. Hybrid search typically improves result quality by capturing both. When disabled, only semantic search is used.

Lexical weight

Relative weight given to lexical (keyword-based) search results when hybrid search is enabled. Increase this weight if users frequently search for specific IDs, error codes, or technical terms. The semantic and lexical weights should typically sum to 1.0.

Semantic weight

Relative weight given to semantic (embedding-based) search results when hybrid search is enabled. Increase this weight if users primarily ask conceptual 'why' or 'how' questions. The semantic and lexical weights should typically sum to 1.0.

Context Assembly

Hierarchical ordering: When enabled, context sections are assembled in a fixed order defined by the priority order setting. LLMs often pay the most attention to the beginning and end of a prompt, so section ordering can influence which information the model prioritizes.
Priority order: Defines the ordering of context sections in the assembled prompt. Sections listed first receive priority when the total token budget is exceeded. LLMs often pay the most attention to the beginning and end of a prompt, so the order can influence response quality. Changing the priority order can help the model focus on the most critical data. Example prompt: "Always put the System Prompt first, then the most relevant docs, then user history". Priority order only applies when hierarchical ordering is enabled.
Include source citations: When enabled, the context package includes citation metadata that links each piece of retrieved content back to its source document. This enables the LLM to attribute claims to specific sources in its response. Default: Enabled

Performance

These values are the infrastructure limits for the Context Manager service itself.

Max Concurrent Requests: Limits how many context assembly requests can execute simultaneously. Protects downstream services from overload. If you see 429 Too Many Requests errors, consider decreasing this value to match your database's capacity.
Request timeout (seconds): Maximum time allowed for a single context assembly request. If retrieval and assembly exceed this limit, the request is canceled and returns an error. If you receive 504 Gateway Timeout errors, increase this value.

Token limits

Max context tokens: Token budget for retrieved documents and data within the context package. If set too low, the AI will not have enough source material to answer accurately. Ensure this value plus the system and user token budgets does not exceed the maximum total tokens. Minimum: 1 token
Max system tokens: Token budget for the system prompt that defines the AI's role, tone, and constraints. If set too low, the AI might ignore its formatting instructions or behavioral guidelines. Ensure this value plus the context and user token budgets does not exceed the maximum total tokens. Minimum: 1 token
Max total tokens: The maximum number of tokens for the entire prompt sent to the LLM, including system, context, and user sections. Higher values allow more documents and detailed instructions but increase latency and cost. Lower values are faster and cheaper, but the model can lose context or miss key facts because they were truncated. Minimum: 1 token
Max user tokens: Token budget for the user's query and conversation history within the context package. Longer user messages can be truncated to fit this limit. Ensure this value plus the system and context token budgets does not exceed the maximum total tokens. Minimum: 1 token
Output token reserve: Number of tokens reserved for the LLM's response generation. The system subtracts this amount from the model's context window before allocating tokens for input context. Higher values enable longer responses but reduce the budget available for retrieved documents. Default: 4,096 tokens Minimum: 256 tokens
Tokenizer safety margin: Safety factor applied to the computed input token budget to compensate for tokenizer mismatch. The system uses the GPT-4 tokenizer for all models, which can under-count tokens for non-OpenAI models. A value of 0.90 means the system reserves 10 percent of the budget as a safety margin. Default: 0.90 Range: 0.5-1.0

Snippets

Max snippets per document: Limits how many text snippets are extracted from each retrieved document. Lower values prevent any single document from dominating the context at the expense of other sources. Higher values provide broader coverage of a document's content but might crowd out other documents and consume more of the token budget. Minimum: 1 snippet
Snippet min length: Minimum character length for an extracted snippet. Snippets shorter than this are discarded. Smaller snippets are more precise but might lack surrounding context needed for the AI to interpret them correctly. Minimum: 1 character
Snippet max length: Minimum character length for an extracted snippet. Snippets shorter than this are discarded. Smaller snippets are more precise but might lack surrounding context needed for the AI to interpret them correctly. Minimum: 1 character

Grounding Scores

These scores (usually 0.0 to 1.0) represent how well the AI's response is supported by the retrieved documents.

Minimum grounding score: The lowest acceptable confidence score for considering a retrieved document. Documents between this value and the required score are included but ranked lower. A lower threshold ensures the user gets an answer more often but increases the risk of unfounded claims.
Required grounding score: The minimum confidence score a retrieved document must achieve to be included in the context. Documents below this threshold are excluded entirely. A higher threshold increases accuracy and reduces hallucinations, but if set too high the system can refuse to answer valid questions because the retrieval confidence is not high enough. A lower threshold ensures answers more often but increases the risk of unfounded claims.

Workflow Execution tab

Agentic Timeout: Maximum execution time for agentic (AI-driven) activities within a workflow. Agentic activities involve multi-step LLM reasoning and can take longer than simple scripts. If the limit is exceeded, the activity is terminated. Default: 300 seconds / 5 minutes Minimum: 1 second.
Max loop iterations: A safety limit that prevents runaway loop execution inside workflows. If a loop node exceeds this number of iterations, the workflow engine terminates it. Default: 10,000 Minimum: 1
Max prompt length: Maximum character length for prompts submitted to agentic workflow activities. Prevents excessively large inputs from causing LLM timeouts or excessive costs. Prompts exceeding this limit are rejected before execution. Default: 100,000 characters / 100 KB Minimum: 1,000 characters.
Script Timeout: Maximum execution time for script activities within a workflow. If a script exceeds this timeout, it is terminated and the activity fails. Default: 300 seconds / 5 minutes Minimum: 1 second.
Script max output (KB): Maximum kilobytes of stdout or stderr captured from a script activity. This limit applies independently to each stream. The system discards output beyond this limit to prevent worker memory exhaustion. The script continues running until it exits or the timeout fires. Default: 1,024 KB (1 MB) Range: 256-2,048 KB
AAP timeout (seconds): Maximum execution time for Ansible Automation Platform job template and workflow template activities within a workflow. If a job exceeds this timeout, the activity is terminated. Default: 3,600 seconds (1 hour) Minimum: 1 second
Approval decision window (seconds): Default time window the approver has to respond to an approval request. Can be overridden per node. If the approver does not respond within this period, the approval expires. Default: 86,400 seconds (24 hours) Minimum: 1 second
Continue on failure (default): Default continue-on-failure behavior for all nodes that support it, including executor nodes, loop, converge, and approval. When enabled, downstream nodes continue executing even if this node fails. Per-node settings override this default. Default: Disabled
Converge wait duration (seconds): Default time a converge node waits for incoming branches to arrive. Can be overridden per node. When this duration expires, the converge node stops waiting and the workflow continues according to the node's continue-on-failure setting. Default: 86,400 seconds (24 hours) Minimum: 1 second
HTTP request timeout (seconds): Maximum execution time for HTTP request activities within a workflow. If a request does not complete within this period, the activity is terminated. Default: 30 seconds Minimum: 1 second
Max wait duration (seconds): Maximum total duration allowed for wait nodes in workflows. If a wait node's configured duration exceeds this limit, the activity fails. Default: 2,592,000 seconds (30 days). Minimum: 1 second
Default retry backoff coefficient: Multiplier applied to the retry interval on each attempt when using exponential backoff. A coefficient of 2.0 doubles the wait time each retry until the wait time reaches the max interval. Default: 2.0 Minimum: 1.0
Default retry initial interval (seconds): Initial wait time before the first retry. Subsequent retries scale this up by the backoff coefficient. Default: 1 second Minimum: 1 second
Default retry max interval (seconds): Maximum wait time between retries. Caps exponential growth so retries never wait longer than this value. Default: 60 seconds Minimum: 1 second
Default retry max retries: Number of retries after the initial attempt for nodes with retry enabled. Set to 0 to disable retries. Applies to executor and approval nodes when no per-node retry policy is set. Default: 3 Minimum: 0

Application tab

Overwrite existing files: Controls whether the system overwrites an existing converted file if one already exists at the target location. When disabled, the system skips conversion if the output file is already present. Default: Disabled
Conversion timeout (seconds): Maximum time allowed for a document conversion operation. If conversion exceeds this limit, it is cancelled. Large or complex documents might require a higher value. Default: 30 seconds Range: 1-300

Authentication tab

Local login: Controls whether non-built-in local users can log in with a password. When disabled, only built-in accounts (such as admin) can authenticate with a password. This setting does not affect identity provider users. Disable this setting after you configure identity providers if local user login is no longer needed. Default: Enabled
Credential maximum lifetime (days): The maximum number of days a service account credential remains valid. When you create or renew a credential, its expiration is capped to this value. Set to 0 so that credentials never expire. Default: 180. Range: 0–730.

Integrations tab

Health check interval: Interval in seconds between automatic integration health checks. The system validates each configured integration on this schedule to detect connectivity or authentication issues. Default: 300 seconds (5 minutes). Minimum: 60 seconds
Connection test timeout: Maximum time in seconds to wait for an integration connection test to complete before marking it as failed. Default: 10 seconds Minimum: 1 second
Health check batch size: Maximum number of integrations validated per health-check run. Higher values validate more integrations per run but increase load. Remaining integrations are picked up on the next run. Default: 500 Minimum: 1
Resource discovery interval: Interval in seconds between automatic integration resource discovery runs. The system re-discovers and syncs each integration's tools and models on this schedule so new resources appear without a manual refresh. Default: 900 seconds (15 minutes) Minimum: 60 seconds
Resource discovery batch size: Maximum number of integrations whose resources are re-discovered per run. Remaining integrations are picked up on the next run. Default: 500 Minimum: 1

Rate Limiting tab

Requests per window: Maximum number of API requests a single user can make within the configured time window. Set to 0 to disable rate limiting. Default: 0 (disabled) Range: 0-10,000
Window duration (seconds): Duration of the rate limiting time window in seconds. The request allowance refills continuously over this interval. Default: 60 seconds. Range: 1-86,400